An automated method for texture segmentation (11) of geophysical data volumes, where texture is defined by double-window statistics of data values, the statistics being generated by a smaller pattern window moving around within a larger sampling window (12). A measure of “distance” between two locations is selected based on similarity between the double-window statistics from sampling windows centered at the two locations (13). Clustering of locations is then based on distance proximity (14).
|
1. A computer-implemented method for performing texture segmentation of one or more volumes of geophysical data, comprising using local statistical distributions of data values, said distributions computed using two moving windows of user-selected size and shape, one being a pattern window and the other a sampling window larger than the pattern window, wherein the pattern window moves about within the sampling window to generate the local statistical distribution for the sampling window's location, and then selecting a distance measure of similarity followed by partitioning the geophysical data into clusters on a basis of application of the distance measure of similarity to the local statistical distributions for pairs of locations.
10. A computer program product, comprising a non-transitory computer usable medium having a computer readable program code embodied therein, said computer readable program code adapted to be executed to implement a method for performing texture segmentation of one or more volumes of geophysical data, said method comprising using local statistical distributions of data values, said distributions computed using two moving windows of user-selected size and shape, one being a pattern window and the other a sampling window larger than the pattern window, wherein the pattern window moves about within the sampling window to generate the local statistical distribution for the sampling window's location, and then selecting a distance measure of similarity followed by partitioning the geophysical data into clusters on a basis of application of the distance measure of similarity to the local statistical distributions for pairs of locations.
2. The method of
3. The method of
5. The method of
6. The method of
7. The method of
8. The method of
9. The method of
11. A method for producing hydrocarbons from a subsurface region, comprising:
conducting a geophysical survey of the subsurface region;
performing texture segmentation on a volume of geophysical data from the geophysical survey, using a method of
relating the clusters to subsurface structure or geo-bodies;
upon identifying subsurface structure or a geo-body having hydrocarbon potential, drilling a well into the subsurface structure or geo-body; and
producing hydrocarbons from the well.
|
This application claims the benefit of U.S. Provisional Patent Application No. 61/453,809, filed Mar. 17, 2011, entitled GEOPHYSICAL DATA TEXTURE SEGMENTATION USING DOUBLE-WINDOWED CLUSTERING ANALYSIS, the entirety of which is incorporated by reference herein.
This invention relates generally to the field of geophysical prospecting and, more particularly, to identification of geobodies in volumes of geophysical data such as seismic data. Specifically, the invention is a method for texture segmentation using double-windowed clustering analysis.
Texture segmentation is a widely studied area of research in signal/image processing, which aims to divide a time-series or spatial dataset into sub-regions that have high internal homogeneity but differ significantly from each other. The seismic analysis community has also produced some methods that can segment seismic data automatically to facilitate faster analysis. However, most of these methods are based on designing “texture attributes”, or derived properties of the signal, i.e. the data, that can highlight different textures a priori. Other methods begin with small homogeneous regions that are then sequentially merged according to some pre-defined similarity criterion. Still other approaches use libraries of templates or training data for segmentation. Some examples of existing approaches are shown in the following list:
Other examples are compiled and discussed in the survey article by Jean-Pascal Aribot entitled Texture Segmentation that may be downloaded at http://www.mics.ch/SumIntU03/JPAribot.pdf.
The present invention is a method that does not require any training data, attributes or pre-defined similarity measures designed to distinguish textures. Instead, it extracts statistical distributions of user-defined windows on the data, and clusters the distributions according to a standard similarity metric between probability distributions, including similarity metrics commonly used in the literature. This lack of pre-defined criteria, combined with the sensitivity of the method, which enables significantly better and more robust results than alternatives, make the present inventive method particularly advantageous for datasets that do not have a well known pre-defined structure, which is often the case for complex datasets such as seismic data.
In one embodiment, the invention is a computer-implemented method for performing texture segmentation of one or more volumes of geophysical data, comprising using local statistical distributions of data values, said distributions computed using two moving windows of user-selected size and shape, one being a pattern window and the other a sampling window larger than the pattern window, wherein the pattern window moves about within the sampling window to generate the local statistical distribution for the sampling window's location, and then selecting a distance measure of similarity followed by partitioning the geophysical data into clusters on a basis of application of the distance measure of similarity to the local statistical distributions for pairs of locations.
The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fees.
The present invention and its advantages will be better understood by referring to the following detailed description and the attached drawings in which:
The invention will be described in connection with example embodiments. However, to the extent that the following detailed description is specific to a particular embodiment or a particular use of the invention, this is intended to be illustrative only, and is not to be construed as limiting the scope of the invention. On the contrary, it is intended to cover all alternatives, modifications and equivalents that may be included within the scope of the invention, as defined by the appended claims. Persons skilled in the technical field will readily recognize that in practical applications of the present inventive method, it is performed on a computer programmed in accordance with the teachings herein. Indeed, the method was developed with automated use in mind, which is one of its advantages.
Texture segmentation (see the references listed in the “Background” section), in one of its many equivalent forms, is a long studied topic in many fields of research involving spatial or time-series data analysis, including but not limited to computer vision, signal processing (telecommunications by wireless/data or optical networks, image processing (e.g., terrestrial, satellite, radar, medical, or robotic), time-series analysis (stock or financial data) and many others. The basic task consists of breaking up a sequence of data values such as in a time-series, or a spatial configuration of values such as in an image, into regions which are internally statistically homogeneous, with clear and quantifiable differences across regions, as shown in
The distinct regions in these figures can then be considered distinct textures, a term that is widely understood by illustration, but without a clear mathematical definition so far either in the theoretical or practice-oriented literature. While there is a lot of literature based on the concept of “textons”, which are elementary texture patterns that can compose repeating image patterns, it is not clear how textons can be defined to be robust or comprehensive. In this document, an alternative intuitive definition of texture is proposed that naturally enables the segmentation (separation) of such datasets into the distinct regions that are obvious to the human observer. With this introduction, the new concept of texture and how it may be used for segmentation follows next.
Definition of Texture as Window Statistics
Consider the following general definition of a multi-dimensional dataset: At each point xiεRN: i=1, . . . K in N-dimensional space, a vector viεRM: i=1, . . . K, is given. For example, a 256×256 RGB-color image has N=2, M=3, K=256×256=65536. In another example, time series data of an industrial process unit with 50 measurement points that produce a single reading each in one minute intervals for an hour would have N=1, M=50, K=60.
Next, define two nested windows of user-defined size and shape on such a dataset. The smaller window, known as the Pattern Window is contained within the larger Sampling Window as shown symbolically in
For example, in
Note that the above definition thus characterizes texture at a spatial scale characterized by the pattern window, with statistical estimation determined by the sampling window. The set of distributions obtained by scaling the two windows can then be viewed as textures at each location at those scales. It is also straight-forward to extend this method to data in any number of dimensions, with any number of independent attributes at each point in that space, i.e. xεRN, y(x)εRM. In this case, with a cuboidal window of size vector w={wk, k=1, . . . N}, the pattern vector V(x) would be the concatenation of all the y's in the window, and would hence have dimension given by
This clearly implies the explosion of the dimensionality of the algorithm, and computational limitations will rule out use of very large window sizes, unless significant parallel computing can be brought to bear. This can be accomplished by the use of Summed Area Tables (SATs), which are standard data structures used in image processing and other spatial data applications like radar, hyper-spectral imaging etc. See, for example, the online reference: Hensley, J., Scheuermann, T., Coombe, G., Singh, M. and Lastra, A. (2005), Fast Summed-Area Table Generation and its Applications, Computer Graphics Forum 24, pp. 547-555.
Next it is shown how the above definitions can be used to perform texture segmentation.
Texture Segmentation Algorithm
In one embodiment of the present invention, a texture segmentation algorithm may be constructed as described in this section. For a given set of window shapes and sizes, compute the distributions px,y(V) of the pattern vectors at each location x,y, using a 2-D data set as an example. This may be considered to be a single two-dimensional distribution, a joint distribution of the pattern window using all its samples within the sampling window. In the Gaussian case this reduces to computing the mean vector and covariance matrix of V at each location. (See PCT Publication No. WO2010/056424). While this looks computationally very daunting, we will discuss a relatively efficient method later that can exploit the high level of redundancy in the computations using the data structure referred to above, i.e. Summed Area Tables (“SATs”).
We then perform a clustering of similar distributions using any standard clustering algorithm, such as K-means. See, for example, http://www-users.cs.umn.edu/˜kumar/dmbook/ch8.pdf. For greater robustness, and to obtain a better handle over the number of clusters, we may also implement a connectivity-based clustering algorithm, that provides a trade-off curve of the number of clusters vs. clustering quality. Such an algorithm is based on a well-known procedure called “agglomerative clustering”, but it maintains clustering metrics that can provide the trade-off curve. See U.S. Patent Publication No. 2010/0121851. Agglomerative clustering and the trade-off curve are illustrated in
Note on the Distance Function Between Distributions
An important ingredient in this process is the distance/similarity metric, which determines which textures are similar based on their distributions p(V) and q(V), i.e. one x,y location having a distribution p(V) and a second x′,y′ location having a distribution q(V). A standard metric used in similar situations is the well known Kullback-Leibler Divergence (K-L) defined as
See M. Bocquest, C. A. Fires, L. Wu, Beyond Gaussian statistical modeling in geophysical data assimilation (http://idl.ul.pt/carlospires/nongaussianity_review-24032010.pdf, pp. 1-51). For multi-dimensional Gaussians, this takes the form
Although widely used, the K-L divergence has some drawbacks, including lack of symmetry. Instead, a preferred embodiment of the present inventive method uses a new distance metric of the form
Apart from having the desirable property of being symmetric in the exchange of p and q, it can be shown that this metric provides more reliable notion of the difference between the distributions, which is explained herein below. Further, the K-L divergences can be shown to be slopes of the function Φ(t) at t=0 and t=1, i.e.
All these observations are depicted in
In particular, the example distributions in
Again, for the Gaussian case, this new distance measure of the present invention can be obtained as follows:
The minimization over t can be solved by using Golden Section Search (see Wikipedia, http://en.wikipedia.org/wiki/Golden_section_search), or directly by solving the following first-order derivative condition by bi-section search.
To summarize, basic steps in one embodiment of the present inventive method that assumes Gaussian statistics are shown in the flowchart of
At step 11, the user considers a multidimensional geophysical data set, and selects a segmentation algorithm that utilizes the double-windowed texture definition of the present invention, i.e. local statistical distributions computed using two moving windows of user-selected size and shape: (1) a pattern window (2) a sampling window larger than the pattern window. At step 12, for each sampling window contained within the data, the statistical distribution (e.g. mean and covariance matrix in simplest case of Gaussian approximation) of all pattern windows contained within the sampling window is computed. At step 13, distances or dissimilarity measures between pairs of distributions are determined (optionally only within a restricted neighborhood). At step 14, distributions are clustered based on distances/dissimilarities. (In the field of data mining, the terms distance, similarity, and dissimilarity can be used interchangeably and in combination, often as modifiers of measure.) At step 15, spatially contiguous clusters are identified as texture segments.
For more detail on steps 12-14, consider a 2-D data set and at each location x,y, consider the sample window centered there. Each discrete cell within the sample window will have a data value associated with it. Next, construct a pattern vector for each possible location of the pattern window within that sample window by concatenating the data values contained in that pattern window into a wx×wy component vector. The distribution of the pattern vectors within the sample window is then computed. For example, this may be done by computing the mean vector and the covariance matrix. Then, this process may be repeated for all other x,y locations represented by the data set. Whether two such locations should be clustered together is determined by applying some similarity/distance measure to their respective pattern vector distributions.
In all the examples below, the various possible choices are restricted to (a) square windows of small/moderate size (b) Multi-variate Gaussian distributions (c) a simple weighted Euclidean distance metric as an approximation to the K-L divergence or its enhancement discussed earlier (d) K-means clustering. These choices are not intended to limit the scope of the invention and were made primarily for simplicity and computational tractability, as only a simple serial MATLAB implementation on a Dell 630 laptop were used. In a more commercial application, the present inventive method might be implemented, for example, on a GPU cluster, with a menu of choices of window sizes, distance metrics and clustering algorithms. We begin with the simple 1-D signal containing 3 distinct textures.
Test Examples of the Present Inventive Method
The first examples, i.e. those shown in
A synthetic 2-D example is shown in
Finally,
All or part of
The foregoing patent application is directed to particular embodiments of the present invention for the purpose of illustrating it. It will be apparent, however, to one skilled in the art, that many modifications and variations to the embodiments described herein are possible. All such modifications and variations are intended to be within the scope of the present invention, as defined in the appended claims.
Patent | Priority | Assignee | Title |
10121096, | Jul 29 2016 | International Business Machines Corporation | Steering seismic texture analysis algorithms using expert input |
11170215, | Apr 28 2016 | Reality Analytics, Inc. | System and method for discriminating and demarcating targets of interest in a physical scene |
Patent | Priority | Assignee | Title |
4916615, | Jul 14 1986 | Conoco Inc. | Method for stratigraphic correlation and reflection character analysis of setsmic signals |
4992995, | Oct 24 1989 | Amoco Corporation | Methods for attenuating noise in seismic data |
5047991, | Apr 28 1989 | Schlumberger Technology Corporation | Lithology identification using sonic data |
5136551, | Mar 23 1989 | System for evaluation of velocities of acoustical energy of sedimentary rocks | |
5265192, | Sep 20 1990 | Atlantic Richfield Company | Method for the automated editing of seismic traces using an adaptive network |
5274714, | Jun 04 1990 | DATRON ADVANCED TECHNOLOGIES, INC | Method and apparatus for determining and organizing feature vectors for neural network recognition |
5416750, | Mar 25 1994 | Landmark Graphics Corporation | Bayesian sequential indicator simulation of lithology from seismic data |
5444619, | Sep 27 1993 | Schlumberger Technology Corporation | System and method of predicting reservoir properties |
5465308, | Jun 04 1990 | DATRON ADVANCED TECHNOLOGIES, INC | Pattern recognition system |
5539704, | Jun 23 1995 | Western Atlas International, Inc. | Bayesian sequential Gaussian simulation of lithology with non-linear data |
5553704, | Oct 11 1994 | International Paper | Stackable can carrier |
5586082, | Mar 02 1995 | TRUSTEES OF COLUMBIA UNIVERSITY, THE, IN THE CITY OF NEW YORK | Method for identifying subsurface fluid migration and drainage pathways in and among oil and gas reservoirs using 3-D and 4-D seismic imaging |
5852588, | Jul 07 1994 | Schlumberger Technology Corporation | Method of processing seismic data |
5940777, | Sep 19 1995 | Elf Aquitaine Production | Automatic seismic pattern recognition method |
6052650, | Feb 27 1997 | Schlumberger Technology Corporation | Enforcing consistency in geoscience models |
6226596, | Oct 27 1999 | Marathon Oil Company | Method for analyzing and classifying three dimensional seismic information |
6236942, | Sep 15 1998 | BUSH SEISMIC TECHNOLOGIES, LLC | System and method for delineating spatially dependent objects, such as hydrocarbon accumulations from seismic data |
6295504, | Oct 25 1999 | Halliburton Energy Services, Inc.; Elf Exploration Production | Multi-resolution graph-based clustering |
6363327, | May 02 2000 | Chroma Graphics, Inc. | Method and apparatus for extracting selected feature information and classifying heterogeneous regions of N-dimensional spatial data |
6411903, | Sep 15 1998 | BUSH SEISMIC TECHNOLOGIES, LLC | System and method for delineating spatially dependent objects, such as hydrocarbon accumulations from seismic data |
6473696, | Nov 04 1999 | ConocoPhillips Company | Method and process for prediction of subsurface fluid and rock pressures in the earth |
6560540, | Sep 29 2000 | ExxonMobil Upstream Research Company | Method for mapping seismic attributes using neural networks |
6574566, | Dec 27 1999 | ConocoPhillips Company | Automated feature identification in data displays |
6618678, | May 26 2000 | CGG SERVICES NL B V | Method of joint analysis and interpretation of the subsurface from multiple seismic derived layer property data sets |
6625541, | Jun 12 2000 | Schlumberger Technology Corporation | Methods for downhole waveform tracking and sonic labeling |
6631212, | Sep 13 1999 | Intellectual Ventures Fund 83 LLC | Twostage scheme for texture segmentation based on clustering using a first set of features and refinement using a second set of features |
6725163, | Sep 10 1999 | Henning Trappe | Method for processing seismic measured data with a neuronal network |
6735526, | Sep 11 1998 | Statoil Petroleum AS | Method of combining directional seismic attributes using a supervised learning approach |
6757614, | Dec 18 2000 | Schlumberger Technology Corporation | Seismic signal processing method and apparatus for generating correlation spectral volumes to determine geologic features |
6804609, | Apr 14 2003 | ConocoPhillips Company | Property prediction using residual stepwise regression |
6847895, | Sep 11 2000 | Institute Francais du Petrole | Method for facilitating recognition of objects, notably geologic objects, by means of a discriminant analysis technique |
6950786, | Oct 10 2000 | Schlumberger Technology Corporation | Method and apparatus for generating a cross plot in attribute space from a plurality of attribute data sets and generating a class data set from the cross plot |
6957146, | Dec 24 2001 | PGS Geophysical AS | System for utilizing seismic data to estimate subsurface lithology |
6970397, | Jul 09 2003 | Gas Technology Institute | Determination of fluid properties of earth formations using stochastic inversion |
7039239, | Feb 07 2002 | Monument Peak Ventures, LLC | Method for image region classification using unsupervised and supervised learning |
7095677, | May 27 2003 | ASPEN PARADIGM HOLDING LLC | Crossplot analysis of A.V.O. anomolies in seismic surveying |
7162463, | Jul 12 2002 | IKON SCIENCE AMERICAS, INC | Pattern recognition template construction applied to oil exploration and production |
7184991, | Jul 12 2002 | IKON SCIENCE AMERICAS, INC | Pattern recognition applied to oil exploration and production |
7188092, | Jul 12 2002 | IKON SCIENCE AMERICAS, INC | Pattern recognition template application applied to oil exploration and production |
7248258, | Sep 17 2001 | Landmark Graphics Corporation | System and method for analyzing and imaging three-dimensional volume data sets |
7295706, | Jul 12 2002 | CHROMA GROUP, INC | Pattern recognition applied to graphic imaging |
7308139, | Jul 12 2002 | IKON SCIENCE AMERICAS, INC | Method, system, and apparatus for color representation of seismic data and associated measurements |
7453767, | Nov 25 2003 | Method for deriving a 3D GRAZ seismic attribute file | |
7463552, | Nov 25 2003 | Method for deriving 3D output volumes using filters derived from flat spot direction vectors | |
7668043, | Oct 20 2004 | Schlumberger Technology Corporation | Methods and systems for sonic log processing |
20030200030, | |||
20040190374, | |||
20040210394, | |||
20040240321, | |||
20040260476, | |||
20050043892, | |||
20050125157, | |||
20050171700, | |||
20050222774, | |||
20060100785, | |||
20060115145, | |||
20080212841, | |||
20100121851, | |||
WO184462, | |||
WO2010056424, | |||
WO9964896, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Feb 29 2012 | ExxonMobile Upstream Research Company | (assignment on the face of the patent) | / |
Date | Maintenance Fee Events |
Aug 21 2018 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Sep 06 2022 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Date | Maintenance Schedule |
Mar 17 2018 | 4 years fee payment window open |
Sep 17 2018 | 6 months grace period start (w surcharge) |
Mar 17 2019 | patent expiry (for year 4) |
Mar 17 2021 | 2 years to revive unintentionally abandoned end. (for year 4) |
Mar 17 2022 | 8 years fee payment window open |
Sep 17 2022 | 6 months grace period start (w surcharge) |
Mar 17 2023 | patent expiry (for year 8) |
Mar 17 2025 | 2 years to revive unintentionally abandoned end. (for year 8) |
Mar 17 2026 | 12 years fee payment window open |
Sep 17 2026 | 6 months grace period start (w surcharge) |
Mar 17 2027 | patent expiry (for year 12) |
Mar 17 2029 | 2 years to revive unintentionally abandoned end. (for year 12) |